Our founding thesis: an on-field event changes a game's win probability before the Kalshi market reprices to match it, and the gap then closes. We logged it all season, then instrumented the order book to test it directly. It did not survive the test. Here is the hypothesis, the measurement, and the retraction.
In August 2026 we began capturing Kalshi order-book snapshots around every signal: top-of-book price, spread and depth at the moment of the alert and at intervals out to five minutes afterward. If the correction story were right, the alerted side should drift toward the model's number and the book should thin as others noticed.
Neither happened. Across 81 instrumented MLB signals the median price move on the alerted side was exactly 0.0000¢ at every horizon from t0 to +300 seconds. The median quoted spread stayed flat at 1.0¢. Top-of-book depth rose, from roughly 29.3k to 50.0k contracts, rather than being consumed. There is no measured repricing, no gap closure, and nobody front-running the alert.
The wider record says the same thing. Over our full MLB history of 1,166 priced signals the record is 79.50% at an 80.01¢ average entry, a margin over the entry price of −0.51pp with a 95% interval of [−3.42, +2.29] (p = 0.73): statistically indistinguishable from breakeven, and negative after Kalshi taker fees. Against settled outcomes the model is the badly calibrated party, not the market: on the same game states our static table claimed a 6.47% loss rate, a four-season league table said 12.75%, the market said 20.05%, and reality delivered 20.42%. The market beat the model on Brier score.
So we no longer claim it. Nowhere on this site do we say the line is slow, that the book has not caught up, or that the gap closes. What we do claim is narrower and checkable: we detect the disagreement fast, and we publish every call to a log that is graded at settlement with the losses in it. The rest of this page is preserved as the record of what we believed, why it was plausible, and how the measurement went against it.
This. Fifteen games are live right now and you are watching one of them, which is the entire product: a fast, complete, graded feed of model-versus-price disagreements, sold as an instrument panel rather than as a call to follow. One person can follow one game. Software can follow every live in-game market on Kalshi at once and put a disagreement in your hand about a second after it appears. That advantage is real, it is measurable, and it is the only one we sell.
What we will not sell you is the second half of the old story. We do not tell you the price is about to move, because we instrumented it and it does not. We do tell you, far faster than you would find out yourself, that our model and the market disagree on a game you were not watching, and then we publish what happened, win or lose, whether or not you touched it. Every call is public before you pay us anything. Decide for yourself what that is worth.
A supposed delay between an on-field event changing a game's win probability and the Kalshi market repricing to match it, which would be observable as a gap in percentage points between modeled win probability and the live contract price, and which would close as the market caught up. The gap is real and measurable. The closing is the part our instrumentation did not find.
On Kalshi, a sports market is an event contract: a yes/no contract that settles at $1 if the named team wins and $0 if it doesn't. While the game is live, the price acts like a market-implied win probability — a contract trading at 75¢ means the order book is collectively pricing roughly a 75% chance.
But win probability isn't set by the order book. It's set by the game. A bases-clearing double in the eighth changes the true probability the instant the ball lands. The price only changes when someone actually crosses the spread with an order. Between those two moments, the number on the field and the number on the screen disagree.
That window of disagreement is what we called the Kalshi lag. The disagreement itself is not in doubt: we log both numbers on every signal and they differ, often by a lot. What was in doubt, and is now settled against us, is the second half of the story — that the price would then move toward the model. The alert products and the signals themselves report the disagreement. They do not promise the correction.
Event contracts trade on retail order flow. That single fact drove the whole argument below. It is a good argument. It is also, on our own data, not enough to produce a measurable correction after an alert, which is a useful reminder that a plausible mechanism is not evidence.
Pre-game markets attract most of the volume. Once the first pitch is thrown, resting liquidity thins out — and the fewer resting orders there are, the longer a stale price survives.
Nobody is required to keep in-game quotes current through every pitch. Quotes update when a person — or someone's bot — decides to update them, not on a clock.
See the play, process it, decide, re-price, submit the order. Add broadcast delay on top. The game state changed seconds ago; the book is still catching up.
Walk through the bases-clearing double again with this in mind. The ball lands, three runs score, and the trailing team is suddenly ahead in the eighth. The true win probability just moved by tens of percentage points in about four seconds. On the Kalshi side: some viewers are on a delayed stream, some are watching another game, some see it live but take fifteen seconds to reprice and submit. Until enough of them act, the old price just sits there — quoting a game state that no longer exists.
None of this is a criticism of Kalshi. It's structure, and the reasoning above is why we found the thesis convincing for as long as we did. But the only questions that mattered were how much slower and whether you can measure it — and when we finally measured it properly, the answer was that the alerted side does not move at all. The argument was better than the evidence.
Our MLB model computes win probability from the game state alone — run differential, inning, half, and outs — refreshed roughly every 30 seconds from live game data. In parallel, we pull the live Kalshi price for the same team. When the model's probability exceeds the market price by enough, a signal fires, and we log both numbers at that moment. At settlement, the signal grades as a win or a loss. The full model construction is on the method page, and the anatomy of a signal has a field-by-field breakdown of what gets logged.
The difference at fire — model win probability minus the Kalshi price of the picked side, in percentage points — is what we log. Note what it is: the distance between two estimates, one of which is ours. It is not a measurement of how far the price is about to travel. Across this season's graded MLB dataset:
| Metric | Value |
|---|---|
| Graded signals | 1,382 |
| Graded games (first signal per game) | 949 |
| Median gap at fire | 12.5 pp |
| Mean gap at fire | 13.5 pp |
| Maximum gap at fire | 25 pp |
| Median Kalshi price of picked side at fire | 80¢ |
| First-signal record | 749W–200L · 78.9% at a median entry of 80¢ — about 1 in 5 games is a loss |
Read the median carefully, and read it for what it is: in the typical fired signal, our model and the market's price disagreed by 12.5 percentage points. That is a wide, repeatedly observable disagreement between two estimates. It is not a measurement of how wrong the market was, and given that the market has graded better-calibrated than our model, the safe reading is that a good part of that 12.5 points is our own model's optimism.
BTC works differently. On Kalshi's 15-minute BTC markets our alerts are timed, not triggered by a disagreement: they go out about two minutes before each close whenever the model leans one side at 60% or more. The published record is 91.2% over 12,160 priced signals at a 91.7¢ average entry (96¢ median). Breakeven before fees at that price is 91.7%, and Kalshi's fee pushes it higher. 1,073 of those signals lost. The BTC data has its own studies: what the graded windows show, fees and the breakeven win rate, and how the markets settle.
These are pulled straight from the public log. We show losses for the same reason we show wins: a record you can't audit is marketing.
| Date | Game | Situation | Model | Kalshi at fire | Gap | Final | Result |
|---|---|---|---|---|---|---|---|
| 2026-05-10 | MIN @ CLE | Bottom 11th, one-run game | 100% MIN | 75¢ | 25 pp | MIN 2–1 | WIN |
| 2026-05-21 | NYM @ WSH | Bottom 9th | 90% NYM | 65¢ | 25 pp | NYM 2–1 | WIN |
| 2026-05-30 | DET @ CWS | Bottom 10th | 100% DET | 79¢ | 21 pp | CWS 4–3 (walk-off) | LOSS |
| 2026-06-14 | HOU @ KC | 4th inning | 89% KC | 66¢ | 23 pp | HOU 8–7 (comeback) | LOSS |
The win, up close. May 10, extra innings in Cleveland, a one-run game in the bottom of the 11th. The model read Minnesota at 100% from the game state; Kalshi was quoting MIN at 75¢. That 25-point difference ties the widest in our graded set — a quarter of the contract's entire price range. Minnesota closed out the 2–1 win and the contract settled at $1. Note what this example does not show: it does not show the price climbing toward the model after the alert, and we have no evidence it did. A single win is a single win. The reason to look at it is that the disagreement was real and logged in advance, which is checkable — not that it was money on the table.
May 30, bottom of the 10th on the South Side. The model read Detroit at 100% — a certainty-grade read, the strongest output the model can produce — with Kalshi quoting DET at 79¢. The White Sox then walked it off, 4–3. The signal graded as a loss and went into the public log like every other loss.
Sit with that: even a 100% model read lost. The model computes from run differential, innings, and outs. It does not see bullpen fatigue, who's due up, or a hanging slider. A "100%" read means the historical game states that looked like this one were essentially always converted — not that this one must be. Settlement risk never goes to zero until the contract settles. If someone tells you otherwise about any signal, close the tab.
If the lag story were right, the relationship would be simple: the wider the difference between model and market, the better the signal should grade. Here's what the data actually shows, bucketed by difference at fire across all first-signal games:
| Gap at fire | Games (n) | Win rate | Loss rate |
|---|---|---|---|
| 10–15 pp | 320 | 83.8% | 16.2% |
| 15–20 pp | 82 | 73.2% | 26.8% |
| ≥ 20 pp | 27 | 81.5% | 18.5% |
The 15–20 point bucket — where the market disagrees with the model more — grades meaningfully worse than the 10–15 point bucket, with more than 1 in 4 of those games ending in a loss. That is the wrong way round for the lag story and the right way round for a simpler one: when the market sits that far from our model, it usually knows something the model does not — an injury, a gassed bullpen, weather, a lineup wrinkle — or our model is simply wrong. The order book is not empty-headed.
This is why we publish the buckets instead of hiding them, and why the size of the difference is not a dial you can crank. Read alongside the correction at the top of this page, this table was the first hint of what the order-book data later confirmed: the disagreement is not a queue of money waiting to be collected.
Contracts cost real cents at real prices — the median picked side in our dataset trades at 80¢ when the signal fires. High prices mean small settlement upside per contract and the full stake at risk on every one.
An 78.9% first-signal win rate still means about 1 in 5 games loses. Your entry price, your sizing, and your discipline decide your actual outcome — at 80¢ entries, a single loss costs roughly what four winning contracts return.
200 first-signal losses in 949 games did not show up one at a time on schedule. Losses cluster. If a run of consecutive losses at your stake size would wreck you, the stake is too big — full stop.
This is the one we got wrong for a year, and the reason this page carries a correction. Our own order-book snapshots show a median alerted-side move of 0.0000¢ out to five minutes, a flat spread and rising depth. Whatever the disagreement is, it is not a repricing you can be early to.
On identical game states our model has claimed a 6.47% loss rate where a plain four-season league table said 12.75%, the market said 20.05%, and reality delivered 20.42%. When the model and the market disagree, the market has been the better-calibrated of the two.
One more thing it is not: advice. This page documents a market mechanism. Whether you act on it, on which platform, at what size, or not at all, is entirely yours.
Free, no account, no email. This is the part of the product we still stand behind without qualification. The proof channel at t.me/DegenHedgeProof posts every MLB signal as a locked entry about 10 minutes after it fires, side hidden so it can't tip anyone's hand, then revealed at settlement, losses included, and every NFL signal 12 minutes after it fires with the side shown. It has run continuously since June 13, 2026. BTC and NBA signals are not posted there; they are graded in the results log. Pair it with the public results log — sortable, every loss on display — and the raw CSV download if you'd rather run the numbers yourself. That's the whole verification loop: watch a signal appear mid-game, watch the reveal, check it against the log. You never have to take our word for anything on this page.
Paid, if you want the alerts in real time. The subscription products deliver the same signals to your Telegram the moment they fire, instead of locked-and-delayed. What you are paying for is that speed and that public record, not a claim that you will come out ahead. MLB in-game alerts run through the World Series; NFL in-game alerts cover every game; the BTC 15-minute product uses a different, timed trigger (a read about two minutes before each window closes, around the clock), and NBA alerts are offseason until late October, when both the alerts and the public NBA record open at tip-off. Every feed is $4.99 a month with a 30-day free trial. The full product rundown is on the Kalshi alerts page.
⚾ MLB in-game alerts, delivered on Telegram the moment the model and the market disagree. Every signal graded publicly, wins and losses. Free for 30 days, then $4.99/mo per feed.
Kalshi BTC 15-minute signals on Telegram, each pushed with its live ask about two minutes before the window closes, around the clock. Its published record is 91.2% at a 91.7¢ average entry. Breakeven before fees at that price is the same number, 91.7%. What you are paying for is coverage and speed, not a price advantage. Read the fee and breakeven study first.
Start BTC free, 30 days →/manage in the bot, or email [email protected]. Cancelling later stops future renewals at the end of the current paid month and does not refund it. MLB only: MLB alerts run through the World Series in late October. The feed is quiet over the winter. After the final World Series game we pause billing on MLB subscriptions bought on degenhedge.com, so nothing is charged over the winter. Billing resumes shortly before Opening Day, and we email you at least 7 days before the first charge. Cancel any time from your account page. No refunds except as required by law. By starting a trial you authorize these recurring charges and agree to the Terms and Privacy Policy. Signals are informational and for entertainment only, are not financial or betting advice, and any single signal can lose 100% of your stake. Must be 18+ (some platforms require 21+). If gambling is a problem for you, call 1-800-GAMBLER.